Re: [patch 2.6.13-rc4] fix get_user_pages bug

[Nick Piggin]

Linus Torvalds wrote:

> Instead, I'd suggest changing the logic for "lookup_write". Make it 
> require that the page table entry is _dirty_ (not writable), and then 
> remove the line that says:
>
> 	lookup_write = write && !force;
>
> and you're now done. A successful mm fault for write _should_ always have 
> marked the PTE dirty (and yes, part of testing this would be to verify 
> that this is true - but since architectures that don't have HW dirty 
> bits depend on this anyway, I'm pretty sure it _is_ true).
>
> Ie something like the below (which is totally untested, obviously, but I 
> think conceptually is a lot more correct, and obviously a lot simpler).

Surely this introduces integrity problems when `force` is not set?
Security holes? Perhaps not, but I wouldn't guarantee it.

However: I like your idea. And getting rid of the lookup_write logic is
a good thing.

I don't much like that it changes the semantics of follow_page for
write on a readonly pte, and that is where your problem is introduced.
I think to go down this route you'd at least need a follow_page check
that is distinct from 'write'. 'writeable', maybe.

Then, having a 'writeable' flag lets you neatly "comment" your idea of
what might constitute a writeable pte, as opposed to the current code
which basically looks like black magic to a reader, and gives no indication
of how it satisfies the get_user_pages requirements.

A minor issue: I don't much like the proliferation of __follow_page flags
either. Why not make __follow_page take a bitmask, and be used directly by
get_user_pages, which would also allow removal of the 'write' argument from
follow_page.

I would cook you some patches, but I'm not in front of the source tree.


Send instant messages to your online friends http://au.messenger.yahoo.com 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/